Xianglong Guan, Li Ma, Yunyou Huang, Suqin Tang, Tinghui Li
The process of Alzheimer’s disease (AD) is irreversible, but reasonable medical intervention for preclinical AD can delay AD’s onset. Progressive mild cognitive impairment (pMCI) is the most critical stage for AD preclinical intervention. Therefore, accurate identification of pMCI will significantly improve patient benefits. Functional MRI is a neuroimaging modality that has been widely utilized to study brain activity related to AD. However, it is challenging to obtain functional MRI data, and a small amount of data will easily lead to the overfitting of the identification model. In addition, the current pMCI identification model lack interpretability leads to difficulty in acceptance by clinicians. In this work, we propose an interpretable hybrid model based on a brain network atlas to identify pMCI subjects. First, the hybrid model utilizes multi-layer perceptron to obtain categorical global features to help graph neural networks reduce overfitting. Second, the attention mechanism is introduced into the model to explain the recognition behavior of the model. The results show that our model outperforms the comparison models on multiple metrics.
{"title":"An Interpretable Brain Network Atlas-Based Hybrid Model for Mild Cognitive Impairment Progression Prediction","authors":"Xianglong Guan, Li Ma, Yunyou Huang, Suqin Tang, Tinghui Li","doi":"10.1145/3590003.3590081","DOIUrl":"https://doi.org/10.1145/3590003.3590081","url":null,"abstract":"The process of Alzheimer’s disease (AD) is irreversible, but reasonable medical intervention for preclinical AD can delay AD’s onset. Progressive mild cognitive impairment (pMCI) is the most critical stage for AD preclinical intervention. Therefore, accurate identification of pMCI will significantly improve patient benefits. Functional MRI is a neuroimaging modality that has been widely utilized to study brain activity related to AD. However, it is challenging to obtain functional MRI data, and a small amount of data will easily lead to the overfitting of the identification model. In addition, the current pMCI identification model lack interpretability leads to difficulty in acceptance by clinicians. In this work, we propose an interpretable hybrid model based on a brain network atlas to identify pMCI subjects. First, the hybrid model utilizes multi-layer perceptron to obtain categorical global features to help graph neural networks reduce overfitting. Second, the attention mechanism is introduced into the model to explain the recognition behavior of the model. The results show that our model outperforms the comparison models on multiple metrics.","PeriodicalId":340225,"journal":{"name":"Proceedings of the 2023 2nd Asia Conference on Algorithms, Computing and Machine Learning","volume":"4577 1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114071189","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Abstract: To make up for the deficiencies of the Harris hawk optimization algorithm (HHO) in solving multi-objective optimization problems with low algorithm accuracy, slow rate of convergence, and easily fall into the trap of local optima, a multi-strategy improved multi-objective Harris hawk optimization algorithm with elite opposition-based learning (MO-EMHHO) is proposed. First, the population is initialized by Sobol sequences to increase population diversity. Second, incorporate the elite backward learning strategy to improve population diversity and quality. Further, an external profile maintenance method based on an adaptive grid strategy is proposed to make the solution better contracted to the real Pareto frontier. Subsequently, optimize the update strategy of the original algorithm in a non-linear energy update way to improve the exploration and development of the algorithm. Finally, improving the diversity of the algorithm and the uniformity of the solution set using an adaptive variation strategy based on Gaussian random wandering. Experimental comparison of the multi-objective particle swarm algorithm (MOPSO), multi-objective gray wolf algorithm (MOGWO), and multi-objective Harris Hawk algorithm (MOHHO) on the commonly used benchmark functions shows that the MO-EMHHO outperforms the other compared algorithms in terms of optimization seeking accuracy, convergence speed and stability, and provides a new solution to the multi-objective optimization problem.
{"title":"Multi-strategy Improved Multi-objective Harris Hawk Optimization Algorithm with Elite Opposition-based Learning","authors":"Fulin Tian, Jiayang Wang, Fei Chu, Lin Zhou","doi":"10.1145/3590003.3590030","DOIUrl":"https://doi.org/10.1145/3590003.3590030","url":null,"abstract":"Abstract: To make up for the deficiencies of the Harris hawk optimization algorithm (HHO) in solving multi-objective optimization problems with low algorithm accuracy, slow rate of convergence, and easily fall into the trap of local optima, a multi-strategy improved multi-objective Harris hawk optimization algorithm with elite opposition-based learning (MO-EMHHO) is proposed. First, the population is initialized by Sobol sequences to increase population diversity. Second, incorporate the elite backward learning strategy to improve population diversity and quality. Further, an external profile maintenance method based on an adaptive grid strategy is proposed to make the solution better contracted to the real Pareto frontier. Subsequently, optimize the update strategy of the original algorithm in a non-linear energy update way to improve the exploration and development of the algorithm. Finally, improving the diversity of the algorithm and the uniformity of the solution set using an adaptive variation strategy based on Gaussian random wandering. Experimental comparison of the multi-objective particle swarm algorithm (MOPSO), multi-objective gray wolf algorithm (MOGWO), and multi-objective Harris Hawk algorithm (MOHHO) on the commonly used benchmark functions shows that the MO-EMHHO outperforms the other compared algorithms in terms of optimization seeking accuracy, convergence speed and stability, and provides a new solution to the multi-objective optimization problem.","PeriodicalId":340225,"journal":{"name":"Proceedings of the 2023 2nd Asia Conference on Algorithms, Computing and Machine Learning","volume":"18 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124968430","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Due to the serious interference of illumination and background on the camera during the live operation of the distribution network robot, it is difficult to match, identify, and locate the feature points of the target image, such as the drainage line. This paper proposes the intelligent perception recognition and positioning method of the distribution network drainage line. First, YOLOv4 is used to identify and classify the typical parts of the distribution network and determine the two-dimensional position of the operation point. Subsequently, the Res-Unet segmentation network was improved to perform image segmentation of drainage lines and wires to avoid complex background interference. Finally, binocular vision is used to extract the center line of the wire through the image geometric moment and determine the image line of the wire and the center of the double eyes. The intersection line of the wire is the spatial three-dimensional coordinates of the wire. After the target detection, wire segmentation, and operation point positioning experiments, this method can achieve a positioning accuracy of 1 mm in the x and y directions and 3 mm in the z direction under the camera coordinate system, which provides a guarantee for accurate perception and recognition and reliable operation control of the power distribution robot operation.
{"title":"Intelligent perception recognition and positioning method of distribution network drainage line","authors":"Shuzhou Xiao, Qiuyan Zhang, Q. Fan, Jianrong Wu, Chao Zhao","doi":"10.1145/3590003.3590088","DOIUrl":"https://doi.org/10.1145/3590003.3590088","url":null,"abstract":"Due to the serious interference of illumination and background on the camera during the live operation of the distribution network robot, it is difficult to match, identify, and locate the feature points of the target image, such as the drainage line. This paper proposes the intelligent perception recognition and positioning method of the distribution network drainage line. First, YOLOv4 is used to identify and classify the typical parts of the distribution network and determine the two-dimensional position of the operation point. Subsequently, the Res-Unet segmentation network was improved to perform image segmentation of drainage lines and wires to avoid complex background interference. Finally, binocular vision is used to extract the center line of the wire through the image geometric moment and determine the image line of the wire and the center of the double eyes. The intersection line of the wire is the spatial three-dimensional coordinates of the wire. After the target detection, wire segmentation, and operation point positioning experiments, this method can achieve a positioning accuracy of 1 mm in the x and y directions and 3 mm in the z direction under the camera coordinate system, which provides a guarantee for accurate perception and recognition and reliable operation control of the power distribution robot operation.","PeriodicalId":340225,"journal":{"name":"Proceedings of the 2023 2nd Asia Conference on Algorithms, Computing and Machine Learning","volume":"52 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127102150","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In the video recommendation scenario, knowledge graphs are usually introduced to supplement the data information between videos to achieve information expansion and solve the problems of data sparsity and user cold start. However, there are few high-quality knowledge graphs available in the field of video recommendation, and there are many schemes based on knowledge graph embedding, which have different effects on recommendation performance and bring difficulties to researchers. Based on the streaming media video website data, this paper constructs knowledge graphs of two typical scenarios (i.e., sparse distribution scenarios and dense distribution scenarios ). Moreover, six state-of-the-art knowledge graph embedding methods are analyzed based on extensive experiments from three aspects: data distribution type, data set segmentation method, and recommended quantity range. Comparing the recommendation effect of knowledge graph embedding methods. The experimental results demonstrate that: in the sparse distribution scenario , the recommendation effect using TransE is the best; in the dense distribution scenario, the recommendation effect using TransE or TranD is the best. It provides a reference for subsequent researchers on how to choose knowledge map embedding methods under specific data distribution.
{"title":"Comparative Research on Embedding Methods for Video Knowledge Graph","authors":"Zhihong Zhou, Qiang Xu, Hui Ding, Shengwei Ji","doi":"10.1145/3590003.3590049","DOIUrl":"https://doi.org/10.1145/3590003.3590049","url":null,"abstract":"In the video recommendation scenario, knowledge graphs are usually introduced to supplement the data information between videos to achieve information expansion and solve the problems of data sparsity and user cold start. However, there are few high-quality knowledge graphs available in the field of video recommendation, and there are many schemes based on knowledge graph embedding, which have different effects on recommendation performance and bring difficulties to researchers. Based on the streaming media video website data, this paper constructs knowledge graphs of two typical scenarios (i.e., sparse distribution scenarios and dense distribution scenarios ). Moreover, six state-of-the-art knowledge graph embedding methods are analyzed based on extensive experiments from three aspects: data distribution type, data set segmentation method, and recommended quantity range. Comparing the recommendation effect of knowledge graph embedding methods. The experimental results demonstrate that: in the sparse distribution scenario , the recommendation effect using TransE is the best; in the dense distribution scenario, the recommendation effect using TransE or TranD is the best. It provides a reference for subsequent researchers on how to choose knowledge map embedding methods under specific data distribution.","PeriodicalId":340225,"journal":{"name":"Proceedings of the 2023 2nd Asia Conference on Algorithms, Computing and Machine Learning","volume":"65 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131648880","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The COVID-19 epidemic has been raging overseas for more than three years, and inbound goods and people have become the main risk points of the domestic epidemic. As the main window for China to exchange materials and personnel with foreign countries, under the dual pressure of the global economic downturn and the China-US economic confrontation, ports’ pressure and responsibility to ensure material transportation and foreign trade are particularly heavy. However, the risk screening of ship and crew epidemic information based on manual methods is extremely time-consuming and labor-intensive, and it is difficult to take into account the efficiency and accuracy requirements of the port's own business and disease control and traceability. To this end, this study proposes an epidemic risk screening method based on knowledge graphs. This method is based on shipping big data and community discovery algorithms, analyzes the geospatial similarity of ship information, crew information and real-time epidemic policy information, and quickly establishes a structure. Map data, quickly screen high-risk ships and crew members, and access the business system to arrange nucleic acid testing tasks. When the time cost is only one thousandth of that of manual labor, the detection accuracy rate approaches and exceeds the accuracy level of manual screening, with an average precision advantage of 8.18% and an average time advantage of 1423 times. It is further found that it is more capable of performing heavy screening tasks than humans, and its AUC decline rate with the increase of the amount of measured data is only 34% of that of the manual method. The research results have been initially applied in Ningbo Port, which has greatly improved the informatization level and screening efficiency of Ningbo Port's risk screening during COVID-19 epidemic.
{"title":"Research on Epidemic Big Data Monitoring and Application of Ship Berthing Based on Knowledge Graph-Community Detection","authors":"Dongfang Shang, Yuesong Li, Jiashuai Xu, Kexin Bao, Ruixi Wang, Liu Qin","doi":"10.1145/3590003.3590026","DOIUrl":"https://doi.org/10.1145/3590003.3590026","url":null,"abstract":"The COVID-19 epidemic has been raging overseas for more than three years, and inbound goods and people have become the main risk points of the domestic epidemic. As the main window for China to exchange materials and personnel with foreign countries, under the dual pressure of the global economic downturn and the China-US economic confrontation, ports’ pressure and responsibility to ensure material transportation and foreign trade are particularly heavy. However, the risk screening of ship and crew epidemic information based on manual methods is extremely time-consuming and labor-intensive, and it is difficult to take into account the efficiency and accuracy requirements of the port's own business and disease control and traceability. To this end, this study proposes an epidemic risk screening method based on knowledge graphs. This method is based on shipping big data and community discovery algorithms, analyzes the geospatial similarity of ship information, crew information and real-time epidemic policy information, and quickly establishes a structure. Map data, quickly screen high-risk ships and crew members, and access the business system to arrange nucleic acid testing tasks. When the time cost is only one thousandth of that of manual labor, the detection accuracy rate approaches and exceeds the accuracy level of manual screening, with an average precision advantage of 8.18% and an average time advantage of 1423 times. It is further found that it is more capable of performing heavy screening tasks than humans, and its AUC decline rate with the increase of the amount of measured data is only 34% of that of the manual method. The research results have been initially applied in Ningbo Port, which has greatly improved the informatization level and screening efficiency of Ningbo Port's risk screening during COVID-19 epidemic.","PeriodicalId":340225,"journal":{"name":"Proceedings of the 2023 2nd Asia Conference on Algorithms, Computing and Machine Learning","volume":"21 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130136017","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Visual position and attitude measurement (VPAM) system has been widely used in obtaining space target information. In order to better obtain different target information and meet the requirements, it is particularly important to select a correct and effective measurement algorithm. In this paper, a performance evaluation software of VPAM algorithm is designed, which can compare and analyze the accuracy and complexity of algorithms used by different VPAM models, and help users select appropriate position models to obtain more accurate target information. Finally, the software is verified by using the dual photogrammetric model in the shipborne helicopter landing system, and the validity of the analysis software is verified by comparing the calculation results with the theoretical value of the algorithm accuracy analysis. The main contribution of this paper is that, as far as we know, it is the first time to try to evaluate the complexity and accuracy of the algorithm by building analysis software instead of theoretical analysis.
{"title":"An Analysis Software for Visual Position and Attitude Measurement Algorithm","authors":"Tao-rang Xu, Jing Zhang, Bin Cai, Yafei Wang","doi":"10.1145/3590003.3590043","DOIUrl":"https://doi.org/10.1145/3590003.3590043","url":null,"abstract":"Visual position and attitude measurement (VPAM) system has been widely used in obtaining space target information. In order to better obtain different target information and meet the requirements, it is particularly important to select a correct and effective measurement algorithm. In this paper, a performance evaluation software of VPAM algorithm is designed, which can compare and analyze the accuracy and complexity of algorithms used by different VPAM models, and help users select appropriate position models to obtain more accurate target information. Finally, the software is verified by using the dual photogrammetric model in the shipborne helicopter landing system, and the validity of the analysis software is verified by comparing the calculation results with the theoretical value of the algorithm accuracy analysis. The main contribution of this paper is that, as far as we know, it is the first time to try to evaluate the complexity and accuracy of the algorithm by building analysis software instead of theoretical analysis.","PeriodicalId":340225,"journal":{"name":"Proceedings of the 2023 2nd Asia Conference on Algorithms, Computing and Machine Learning","volume":"36 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130249393","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Medical image segmentation is crucial for facilitating pathology assessment, ensuring reliable diagnosis and monitoring disease progression. Deep-learning models have been extensively applied in automating medical image analysis to reduce human effort. However, the non-transparency of deep-learning models limits their clinical practicality due to the unaffordably high risk of misdiagnosis resulted from the misleading model output. In this paper, we propose a explainability metric as part of the loss function. The proposed explainability metric comes from Class Activation Map(CAM) with learnable weights such that the model can be optimized to achieve desirable balance between segmentation performance and explainability. Experiments found that the proposed model visibly heightened Dice score from to , Jaccard similarity from to and Recall from to respectively compared with U-net. In addition, results make clear that the drawn model outdistances the conventional U-net in terms of explainability performance.
{"title":"Explainable Deep Learning for Medical Image Segmentation With Learnable Class Activation Mapping","authors":"Kaiyu Wang, Sixing Yin, Yining Wang, Shufang Li","doi":"10.1145/3590003.3590040","DOIUrl":"https://doi.org/10.1145/3590003.3590040","url":null,"abstract":"Medical image segmentation is crucial for facilitating pathology assessment, ensuring reliable diagnosis and monitoring disease progression. Deep-learning models have been extensively applied in automating medical image analysis to reduce human effort. However, the non-transparency of deep-learning models limits their clinical practicality due to the unaffordably high risk of misdiagnosis resulted from the misleading model output. In this paper, we propose a explainability metric as part of the loss function. The proposed explainability metric comes from Class Activation Map(CAM) with learnable weights such that the model can be optimized to achieve desirable balance between segmentation performance and explainability. Experiments found that the proposed model visibly heightened Dice score from to , Jaccard similarity from to and Recall from to respectively compared with U-net. In addition, results make clear that the drawn model outdistances the conventional U-net in terms of explainability performance.","PeriodicalId":340225,"journal":{"name":"Proceedings of the 2023 2nd Asia Conference on Algorithms, Computing and Machine Learning","volume":"107 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125002047","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Occluded person re-identification (ReID) is a challenging task in the field of computer vision, facing the problem that the target pedestrians in probe images are obscured by various occlusions. Random Erasing in data augmentation techniques is one of the effective methods used to deal with the occlusion problem, but it may introduce noise into the training process, which affects the training of the model. In order to solve this problem, we propose an novel data augmentation method named Key-retained Random Erasing (KRE) which preserves the critical parts in images for occluded person ReID. Based on the regular Random Erasing, we utilize the naturally generated attention map in Vision Transformers and introduce an adaptive threshold selection method to detect the key areas of the image to be augmented. The complexity of the training samples can be improved without losing the key information of the images by reserving the key areas in Random Erasing process, which can finally alleviate the occluded person ReID problem. Validating the proposed method on occluded, partial and holistic ReID datasets, extensive experimental results demonstrate that our method performs favorably against state-of-the-art methods on ViT-based models.
遮挡人再识别(ReID)是计算机视觉领域的一项具有挑战性的任务，它面临着探测图像中目标行人被各种遮挡遮挡的问题。数据增强技术中的随机擦除是处理遮挡问题的有效方法之一，但它可能会在训练过程中引入噪声，影响模型的训练。为了解决这一问题，我们提出了一种新的数据增强方法——密钥保留随机擦除(Key-retained Random erase, KRE)。在常规随机擦除的基础上，利用视觉变形中自然生成的注意图，引入自适应阈值选择方法来检测待增强图像的关键区域。通过在Random erase过程中保留关键区域，可以在不丢失图像关键信息的情况下提高训练样本的复杂度，最终缓解被遮挡人的ReID问题。在遮挡的、部分的和整体的ReID数据集上验证了所提出的方法，大量的实验结果表明，我们的方法在基于vit的模型上优于最先进的方法。
{"title":"KRE: A Key-retained Random Erasing Method for Occluded Person Re-identification","authors":"Hongxia Wang, Yao Ma, Xiang Chen","doi":"10.1145/3590003.3590089","DOIUrl":"https://doi.org/10.1145/3590003.3590089","url":null,"abstract":"Occluded person re-identification (ReID) is a challenging task in the field of computer vision, facing the problem that the target pedestrians in probe images are obscured by various occlusions. Random Erasing in data augmentation techniques is one of the effective methods used to deal with the occlusion problem, but it may introduce noise into the training process, which affects the training of the model. In order to solve this problem, we propose an novel data augmentation method named Key-retained Random Erasing (KRE) which preserves the critical parts in images for occluded person ReID. Based on the regular Random Erasing, we utilize the naturally generated attention map in Vision Transformers and introduce an adaptive threshold selection method to detect the key areas of the image to be augmented. The complexity of the training samples can be improved without losing the key information of the images by reserving the key areas in Random Erasing process, which can finally alleviate the occluded person ReID problem. Validating the proposed method on occluded, partial and holistic ReID datasets, extensive experimental results demonstrate that our method performs favorably against state-of-the-art methods on ViT-based models.","PeriodicalId":340225,"journal":{"name":"Proceedings of the 2023 2nd Asia Conference on Algorithms, Computing and Machine Learning","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129009280","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In response to the current problem of a large amount of abnormal data in parking behavior detection, this research proposes a network specialized in parking behavior identification, which identifies the background parking behavior data, classifies the data with high accuracy, reduces the cost of manually verifying the data in the background, speeds up the parking charging cycle of enterprises, and optimizes the user experience.The dynamic position embedding is introduced in the parking-transformer species, so that the self-attention within the transformer can dynamically model the structure of the input token and dynamically encode the input parking behavior sequence data to improve the accuracy of the model for parking behavior recognition.In addition, we created a self-collected parking behavior(SPB) dataset, which was acquired in a natural state and contained various behaviors, and manually classified the various behaviors within the data, and then randomly divided into a test set and a validation set for training and testing, respectively.Compared with the existing methods, indicate that parking-trasnformer hits acceptable trade-offs,namely,97.14% accuracy for SPB dataset.
{"title":"End-to-end Parking Behavior Recognition Based on Self-attention Mechanism","authors":"Penghua Li, Dechen Zhu, Qiyun Mou, Yushan Tu, Jinfeng Wu","doi":"10.1145/3590003.3590072","DOIUrl":"https://doi.org/10.1145/3590003.3590072","url":null,"abstract":"In response to the current problem of a large amount of abnormal data in parking behavior detection, this research proposes a network specialized in parking behavior identification, which identifies the background parking behavior data, classifies the data with high accuracy, reduces the cost of manually verifying the data in the background, speeds up the parking charging cycle of enterprises, and optimizes the user experience.The dynamic position embedding is introduced in the parking-transformer species, so that the self-attention within the transformer can dynamically model the structure of the input token and dynamically encode the input parking behavior sequence data to improve the accuracy of the model for parking behavior recognition.In addition, we created a self-collected parking behavior(SPB) dataset, which was acquired in a natural state and contained various behaviors, and manually classified the various behaviors within the data, and then randomly divided into a test set and a validation set for training and testing, respectively.Compared with the existing methods, indicate that parking-trasnformer hits acceptable trade-offs,namely,97.14% accuracy for SPB dataset.","PeriodicalId":340225,"journal":{"name":"Proceedings of the 2023 2nd Asia Conference on Algorithms, Computing and Machine Learning","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129108864","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Many-objective optimization problems (MaOPs), are the most difficult problems to solve when it comes to multiobjective optimization issues (MOPs). MaOPs provide formidable challenges to current multiobjective evolutionary methods such as selection operators, computational cost, visualization of the high-dimensional trade-off front. Removal of the reductant objectives from the original objective set, known as objective reduction, is one of the most significant approaches for MaOPs, which can tackle optimization problems with more than 15 objectives is made feasible by its ability to greatly overcome the challenges of existing multi-objective evolutionary computing techniques. In this study, an objective reduction evolutionary multiobjective algorithm using adaptive density-based clustering is presented for MaOPs. The parameters in the density-based clustering can be adaptively determined by depending on the data samples constructed. Based on the clustering result, the algorithm employs an adaptive strategy for objective aggregation that preserves the structure of the original Pareto front as much as feasible. Finally, the performance of the proposed multiobjective algorithms on benchmarks is thoroughly investigated. The numerical findings and comparisons demonstrate the efficacy and superiority of the suggested multiobjective algorithms and it may be treated as a potential tool for MaOPs.
{"title":"An Objective Reduction Evolutionary Multiobjective Algorithm using Adaptive Density-Based Clustering for Many-objective Optimization Problem","authors":"Mingjing Wang, Long Chen, Huiling Chen","doi":"10.1145/3590003.3590103","DOIUrl":"https://doi.org/10.1145/3590003.3590103","url":null,"abstract":"Many-objective optimization problems (MaOPs), are the most difficult problems to solve when it comes to multiobjective optimization issues (MOPs). MaOPs provide formidable challenges to current multiobjective evolutionary methods such as selection operators, computational cost, visualization of the high-dimensional trade-off front. Removal of the reductant objectives from the original objective set, known as objective reduction, is one of the most significant approaches for MaOPs, which can tackle optimization problems with more than 15 objectives is made feasible by its ability to greatly overcome the challenges of existing multi-objective evolutionary computing techniques. In this study, an objective reduction evolutionary multiobjective algorithm using adaptive density-based clustering is presented for MaOPs. The parameters in the density-based clustering can be adaptively determined by depending on the data samples constructed. Based on the clustering result, the algorithm employs an adaptive strategy for objective aggregation that preserves the structure of the original Pareto front as much as feasible. Finally, the performance of the proposed multiobjective algorithms on benchmarks is thoroughly investigated. The numerical findings and comparisons demonstrate the efficacy and superiority of the suggested multiobjective algorithms and it may be treated as a potential tool for MaOPs.","PeriodicalId":340225,"journal":{"name":"Proceedings of the 2023 2nd Asia Conference on Algorithms, Computing and Machine Learning","volume":"20 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128325571","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：